Siemens and PhysicsX collaborate to build AI-based deep physics simulation

PhysicsX is building its latest pre-trained Large Physics Model (LPM) for aerodynamics on high-fidelity simulation data generated with Siemens' Simcenter simulation software (Image credit: PhysicsX)

Together, they aim to create the next generation of AI-based deep physics simulation to accelerate performance prediction and optimisation. PhysicsX is building its latest pre-trained deep physics model for aerodynamics on high-fidelity simulation data generated with the Siemens Xcelerator portfolio.

"AI represents a truly transformative opportunity for the simulation community, and we are exploring its potential at an accelerated rate," said Jean-Claude Ercolanelli, senior vice president, Simulation and Test Solutions, Siemens Digital Industries Software. "To scale AI adoption, open collaboration is key and Siemens is delighted to see PhysicsX unveil to the world what it is working on - together, we're exploring an AI-enhanced simulation industry that has the potential to change how products are ideated and engineered."

"We are delighted to take our collaboration with Siemens to the next level, working together on fundamental technology to improve the way engineering is practiced," said Robin Tuluie, founder and chairman of PhysicsX. "The basis for successful AI deployment into engineering hinges on highest-quality synthetic data, robust integrations between Computer Aided Engineering (CAE) and AI [Artificial Intelligence] and building on the trust customers have in our respective technologies.

"We are thrilled to be building into the opportunity space together with Siemens with the release of our latest Large Geometry Model (LGM) for aerodynamics, LGM-Aero, trained on high-fidelity data generated with Siemens Xcelerator portfolio tools. We are also delighted to be releasing Ai.rplane, an open-access reference application showcasing the power of LGM-Aero to design innovative aero solutions."

Large Geometry Model (LGM) release

Much of the promise of AI in industrial applications starts with the dramatic acceleration that AI can bring to simulation. PhysicsX's LGM-Aero is trained on a corpus of more than 25 million geometries and associated physics simulation. The training data contains tens of billions of mesh elements and tens of thousands of Computational Fluid Dynamics (CFD) and Finite Element Analysis (FEA) simulations using high-quality Simcenter STAR-CCM+ software and Simcenter Nastran software. It allows users to fine-tune deep learning models for their own applications with as little as a few tens of simulations.

PhysicsX has also released Ai.rplane, a free-to-use public-access reference application offering a set of exploratory tools focused on aerodynamics and aircraft structures with unique AI-shape generation. Powered by LGM-Aero, Ai.rplane showcases the power of a fully trained model that generalises to a broad set of aeroelastic applications. LGM-Aero is used as a zero-shot model to infer aero performance, flight stability and structural stress for any flying shape generated. The application demonstrates the creation of geometry and the assessment of physics performance in less than a second, compared to the several hours required for traditional numerical simulations.

Starting with a basic set of instructions, users can explore the full generative design space offered by Ai.rplane, then modify or optimise for a desired performance characteristic in seconds. Over time, PhysicsX aims to add new features, including cargo packaging, selection of commercial powertrains, and controls.

Ai.rplane was developed using an extensive set of simulation technologies from Siemens to automate the generation of high-quality training data. Both LGM-Aero and Ai.rplane are available on the PhysicsX AI engineering platform.